Making Fractional Distances work in the presence of White Noise - Motivation and Challenges

Kalithasan, Namasivayam and Gupta, Megha and Jayaram, Balasubramaniam (2022) Making Fractional Distances work in the presence of White Noise - Motivation and Challenges. In: 5th ACM India Joint 9th ACM IKDD Conference on Data Science and 27th International Conference on Management of Data, CODS-COMAD 2022, 7 January 2022 through 10 January 2022, Virtual, Online.

[img] Text
ACM_International_Conference_Proceeding_Series.pdf - Published Version
Restricted to Registered users only

Download (819kB) | Request a copy

Abstract

In Similarity Search (SS), given a new piece of data (or a query), often a close enough match to it from a given set of data points is sought. One view of SS is that the query is assumed to be a noise corrupted data point. In line with this view, François et. al. [1] argue that the Euclidean norm and fractional distances give better search results in the case of white noise and highly coloured noise, respectively. Further, Singh and Jayaram [2] showed that the fractional distances work well even when the noise is not-so-highly coloured. In this work, we attempt to determine if fractional distances could be made to work in the setting of SS even when the noise is white. The real challenges lie in the many counter-intuitive phenomena in high dimensional spaces which will be discussed briefly and our approach in tackling them will also be discussed in this work. © 2022 Owner/Author.

[error in script]
IITH Creators:
IITH CreatorsORCiD
Jayaram, Balasubramaniamhttp://orcid.org/0000-0001-7370-3821
Item Type: Conference or Workshop Item (Paper)
Uncontrolled Keywords: Coloured noise; Minkowski distances; Similarity search; Sparse representation; White noise
Subjects: Mathematics
Mathematics > General principles of mathematics
Divisions: Department of Mathematics
Depositing User: . LibTrainee 2021
Date Deposited: 15 Jul 2022 10:47
Last Modified: 15 Jul 2022 10:47
URI: http://raiithold.iith.ac.in/id/eprint/9734
Publisher URL: http://doi.org/10.1145/3493700.3493748
Related URLs:

Actions (login required)

View Item View Item
Statistics for RAIITH ePrint 9734 Statistics for this ePrint Item